Personal intelligence
What changes for one person?Memory, skills or personal weights. These are distinct mechanisms, even when all are described as learning.
Research edition 01
Who owns
the learning?
A competitive landscape of decentralized intelligence and per-user evolving AI. The companies, the capital, and the mechanisms behind the claims.
12Organizations & protocols
4Architecture families
78Linked evidence records
13 Oct ’26Research cutoff & verification date
01 / The landscape
Four different problems.
One crowded vocabulary.
“Decentralized AI” can describe who holds the data, who changes the weights, who provides compute, or who captures the value. Organizations may appear in more than one family.
Federated learning
Can data stay with its owner?Participants train locally and share updates. A central coordinator may remain; privacy needs explicit safeguards.
Shared-model infrastructure
Who trains or verifies the model?Independent machines collaborate on training or auditing a shared model. This is not automatically personal intelligence.
Intelligence ecosystems
Who gets rewarded for useful work?Agents, models and evaluators participate in a market. Incentive coordination need not imply shared weights.
The groupings are analytical, not company certifications or rankings. Pure GPU marketplaces are outside the core scope unless they also contribute substantive learning or verification technology.
Follow the learning signal
01 / Input
New experienceConversations, files or task outcomes.
02 / Update
Memory or skillPersist facts, retrieve context, or save a reusable procedure.
03 / Next use
Better contextThe same base model receives a different prompt or tool.
04 / Key distinction
Weights may stay fixedUseful adaptation does not require fine-tuning.
Examples: Hermes Agent, EvoSkill and memory-centric personal assistants. These are schematic mechanisms; implementation varies by product. [NO7] [SN3] [PA6]
02 / Interactive comparison
Inspect the evidence.
Filter the map, matrix, funding chart, timeline and profiles together. Select up to three companies in the profile cards for a side-by-side dossier.
Showing all 12 organizations
What actually changes?
Cells summarize the reviewed evidence. “Documented” means a source, artifact or explicit implementation record exists; it does not mean we reproduced the system. Follow the company name for the full assessment.
| Organization | Per-user weights | Memory / skills | Multi-party mechanism | Public implementation | What moves between participants? |
|---|---|---|---|---|---|
| Personal AIPersonal intelligence | Claimed | Documented | Not established | Edge trials | No demonstrated cross-user learning loop |
| PIN AIPersonal intelligence | Research / plan | Personal Index | Research / plan | App + code | Agent services and data context; shared learning unproven |
| HylonPersonal intelligence | Proposed LoRA | Assistant claim | Proposed FL | Loop unverified | DP + secure aggregation proposed |
| SELF OSPersonal intelligence | Not shown | Runtime state | Mock boundary | Prototype code | Mesh-related interfaces / mocks |
| PluralisShared-model infrastructure | Shared model | Not core | Research run | Papers / run | Compressed activations + backward signals |
| Nous ResearchShared-model infrastructure | Agent ≠ weights | Memory / skills | Model releases | Models + code | Compressed optimizer updates; separate agent state |
| Prime IntellectShared-model infrastructure | Customer model | Not core | Training runs | Models + tools | Weight deltas or trajectories, depending on stage |
| GensynShared-model infrastructure | Not core | Not core | Audit network | Audit artifacts | Reproducible execution evidence / audit results |
| Flower LabsFederated learning | SDK / example | App-specific | FL framework | Runnable tools | Model / adapter updates; privacy layers optional |
| TensorOpera / FedMLFederated learning | SDK capability | Agent context | FL toolkit | Framework | Deployment-specific model updates |
| SentientIntelligence ecosystems | Skills ≠ weights | Reusable skills | Agent network | Code + papers | Skills / tools / agent services |
| BittensorIntelligence ecosystems | Subnet-specific | Subnet-specific | Incentive net | Protocol docs | Scores / outputs / services; not mandatory gradients |
DocumentedCompany claimReportedRoadmapUnverifiedDifferent mechanism
A public SDK supports a developer-built design; it does not prove that every end user has a private evolving model. “Not shown” means not established by this research, not proven absent.
Capital, with the denominator visible.
USD millions. Disclosed capital spans different periods and levels of completeness. These are neither current cash balances nor comparable valuations. Venture and seed capital are shown separately from token proceeds.
Nous Research$158M
Reported cumulative total, October 2026 · Round ledger ↓
Prime Intellect>$150M
Company cumulative total; lower-bound bar · Round ledger ↓
Sentient$85M
Disclosed seed; separate token entry excluded · Round ledger ↓
Gensyn$67.3M
Calculated disclosed venture rounds; excludes token sale · Round ledger ↓
Flower Labs$23.6M
Sum of two identified rounds; may not be lifetime total · Round ledger ↓
TensorOpera / FedML$13.2M
Undated company seed figure; differs from dated $11.5M · Round ledger ↓
PIN AI$10M
Disclosed pre-seed; not a verified lifetime total · Round ledger ↓
Pluralis$7.6M
Disclosed seed round · Round ledger ↓
Recent evidence, not just future plans.
A selection of dated releases, research and financing events at or before the cutoff. An announcement or paper is not independent confirmation of all performance claims.
- Prime Intellect
Moondream acquisition announced
Extends multimodal models and inference; price undisclosed. [PR4]
- Nous Research
New financing and business expansion
$90M announced in October; Hermes business strategy described. [NO1]
- Gensyn
open-1b audit artifacts released
Centralized original training, decentralized verification. [GE4]
- Personal AI
Comcast AI Grid trials announced
A telecom deployment direction for personal memory. [PA7]
- Sentient
GRID launch reported
Composable agents and models, not proof of personal weight learning. [SN4]
Hylon and SELF OS have no sufficiently substantiated dated milestone included here. Absence from this timeline does not imply inactivity.
No organizations match those filters.
Try a broader term or reset the filters.
03 / Company dossiers
From positioning to implementation.
Each dossier includes business evidence, leadership, technical mechanisms, deployed work and the remaining roadmap.
01 / Personal intelligence
Personal AI
Proprietary personal models, now moving into telecom memory infrastructure.
Funding snapshotTotal unresolvedConflicting issuer figures; no reconciled current total.
MaturityCommercial product + edge trials
Per-user weightsClaimed
Read the dossier
Purpose & product
Company claim Make persistent, identity-specific intelligence available to individuals and organizations through memory-based small models. [PA7]
Company claim Personal AI offers memory and persona products. Its March 2026 release describes Comcast AI Grid trials for voice and personalized inference; trials are not evidence of a nationwide rollout. [PA5] [PA7]
Market & go-to-market
Analysis Professional and enterprise knowledge, customer-facing assistants and telecom services. Go-to-market is expanding from direct product adoption toward operator and infrastructure partnerships. Current audited customer counts and revenue were not found. [PA7]
Founding & leadership
Documented 2020 public founding; Human AI Labs legally incorporated on 20 November 2019. [PA1] [PA2]
- CEO
- Suman Kanuganti ↗Named by issuer [PA7]
- CTO
- Sharon (Xiaoran) Zhang ↗Company / filing evidence [PA1] [PA2]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Company claim $7.8M cumulative seed capital announced January 2023. [PA8]
Company claim Republic campaign closed April 2024. The page shows $1,769,074 in one place and $1,694,075 in closing materials; treat proceeds as unresolved. [PA3]
Reported A filing index lists $14.13M on 18 December 2025. The original filing was not recovered; do not add this to earlier figures or call it a verified Series A. [PA4]
Documented FY2023 revenue: $138,319; net loss: $2,837,808, per filed statements. These historical figures do not establish 2026 revenue, cash or runway. A current cumulative funding total remains unresolved. [PA1]
How the system learns
Company claim Earlier MODEL-2 materials describe separately trained expert models and persona memory banks. Documentation also describes retrieving memory blocks into context. The current edge architecture combines specialist SLMs with a Memory Core. Public documentation does not specify which new interactions change weights versus only memory. [PA5] [PA6] [PA7]
Unverified Personalization is proprietary. Weight export, training cadence, deletion from weights and inter-user federated exchange were not established by the reviewed sources.
Implemented / publicly evidenced
Documented Product documentation covers personas and memory retrieval. [PA5] [PA6]
Company claim Edge inference trials with Comcast were announced in March 2026. Performance and cost claims are issuer-reported. [PA7]
Roadmap / not yet established
Roadmap Broader telecom deployment and persistent multimodal memory are strategic directions. No dated, independently verifiable delivery schedule was found. [PA7]
What to verify next
Ask for a demonstration that separates memory retrieval from weight updates, and for contractual portability and deletion guarantees.
Profile last verified: · 8 cited sources · Official site / repository ↗
02 / Personal intelligence
PIN AI
A personal context network with a research path toward private model adaptation.
Funding snapshot$10MDisclosed pre-seed; not a verified lifetime total
MaturityApp distribution + research
Per-user weightsResearch / planned
Read the dossier
Purpose & product
Company claim Let users control personal context and use an open network of agents to fulfill intents. [PI1]
Documented Current product page offers iOS, Android and APK distribution, with personal-assistant and voice-keyboard positioning. The existence of download links does not validate every claimed capability or app-store availability in every region. [PI5]
Market & go-to-market
Analysis Consumer personal assistants and Web3 users; developer agents supply services. App onboarding, data connectors and community incentives are the acquisition paths. Intent-service economics remain less substantiated than the app surface. [PI1] [PI5]
Founding & leadership
Reported 2024 launch / operating year; exact incorporation date not verified. [PI1]
- CEO
- Not verifiedNo current CEO title established [PI1]
- CTO
- Current title not verifiedRegan Peng was called CTO in 2025; his 2026 bio says president [PI6] [PI7]
- Technical leadership
- Regan Peng ↗Cofounder / president; formerly engineering lead [PI1] [PI6]
- Chief AI scientist
- Bill Sun ↗Title in founding announcement [PI1]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Company claim $10M pre-seed announced 9 September 2024; Hack VC, a16z CSX and other investors. No later primary-source round was reconciled. [PI1]
Unverified Revenue, cash, burn, paid-user count and current valuation were not verified. Token rewards are not evidence of revenue.
How the system learns
Company claim Personal Index organizes user context for retrieval / graph retrieval; hybrid local and cloud inference adapts to device capacity. The platform design puts continuous user-specific fine-tuning in future work. zkTLS supports data provenance; it is not a general proof of learning privacy. [PI2]
Reported Guardian of Data proposes training a private assistant through simulated queries and curriculum learning in trusted execution environments. A research paper and repository exist; mass deployment of continuously evolving per-user weights is not established. [PI3] [PI4]
What to verify next
Request device-level evidence of weight updates, the exact trust boundary of cloud / TEE processing, and retention controls.
Profile last verified: · 7 cited sources · Official site / repository ↗
03 / Personal intelligence
Hylon
The closest stated federated personal-AI vision; implementation evidence is limited.
Funding snapshotUndisclosedNo verified external funding ledger
MaturityAssistant / beta claims; learning roadmap
Per-user weightsPlanned / disputed live claims
Read the dossier
Purpose & product
Company claim Combine private personal models with collective improvement and distributed ownership, particularly for underserved or restricted-access markets. [HY1]
Company claim OrbVPN documentation describes a hybrid support assistant and labels decentralized learning as future work. Hylon’s own pages additionally advertise node beta and describe some device tiers as live. These descriptions are not fully reconciled. [HY1] [HY2] [HY4]
Market & go-to-market
Analysis Existing OrbVPN / OrbNet distribution is the proposed entry point; ORB rewards are intended to recruit contributors. A repeatable, separately measured Hylon AI business is not established. [HY1] [HY2]
Founding & leadership
Unverified Founding date and separate legal entity not verified. OrbVPN history should not be substituted for Hylon’s founding date.
- CEO
- Not verifiedNo attributable current profile found
- CTO
- Not verifiedNo attributable current profile found
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Unverified No verified round dates, amounts or investors found. ORB incentives and treasury plans are not completed fundraising.
Unverified No Hylon-specific revenue, audited user count, cash balance or valuation established. Parent-network distribution claims do not prove active AI users. [HY1]
How the system learns
Roadmap The whitepaper proposes local LoRA adapters and encrypted memory, clipped and noised updates, secure aggregation, and GPU-based global training. Phones primarily handle inference; PCs may adapt models. A gated champion–challenger process would approve global releases. [HY3] [HY4]
Unverified DP and secure aggregation are design claims. Public privacy budgets, attack evaluations and evidence of a complete deployed training loop were not found. The whitepaper’s approximately one-million-device architecture is not a verified active-node count. [HY4]
Implemented / publicly evidenced
Company claim Support-assistant and node-beta claims exist, but independently auditable evidence for per-user adaptation plus global aggregation is missing. [HY1] [HY2]
Roadmap / not yet established
Roadmap Gate 0: self-hosted model and personal adaptation; later gates target a product niche, autonomous improvement and frontier-scale capability. Relative durations lack a dependable start date and are conditional, not calendar commitments. [HY4]
What to verify next
Require a running learning-loop demo, public release artifacts, named accountable leadership and independent privacy validation before treating the architecture as shipped.
Profile last verified: · 4 cited sources · Official site / repository ↗
04 / Personal intelligence
SELF OS
A local-first agent shell; public examples are not a trained personal model.
Funding snapshotUndisclosedNo verified funding ledger
MaturityCommunity prototype / integration shell
Per-user weightsNot demonstrated
Read the dossier
Purpose & product
Company claim Provide a personal operating layer for agents, plugins and user-owned workflows. [SE1]
Documented The community edition supplies mobile shells, agent/runtime interfaces and example integrations. Its README explicitly limits it to deterministic example agents; users supply provider integrations and credentials. [SE1]
Market & go-to-market
Analysis Developer/community distribution through a public repository, with a broader managed ecosystem described beyond the open edition. Commercial conversion and pricing are not verified. [SE1] [SE2]
Founding & leadership
Unverified SELF Labs incorporation and SELF-OS project launch date not reliably established. This profile covers selflabs/SELF-OS only.
- CEO
- Milan Cheeks ↗SELF Labs founder / CEO attribution [SE3] [SE4]
- CTO
- Not verifiedNo current attributable CTO profile found
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Unverified No attributable public venture round, amount or valuation verified for this entity.
Unverified Revenue, paid adoption and operating scale not verified. Repository activity is not a commercial-traction metric.
How the system learns
Documented AgentRuntime and PluginRegistry coordinate actions. HarmonyMesh, wallet, catalog and Aura-related surfaces include mocks or boundaries rather than a complete decentralized production backend. No bundled personal-model training loop is demonstrated. [SE1] [SE2]
Analysis Local orchestration can improve control over workflow state. It does not establish ownership of remote provider weights or independence from cloud inference.
Implemented / publicly evidenced
Documented Public integration shell and deterministic example agents. [SE1]
Roadmap / not yet established
Roadmap The broader managed runtime, mesh and ecosystem described by the project exceed the community edition. No independently substantiated delivery dates were found. [SE2]
What to verify next
Ask which components are executable today, which require managed services, and whether learning changes memory, code or weights.
Profile last verified: · 4 cited sources · Official site / repository ↗
05 / Shared-model infrastructure
Pluralis
Independent parties hold parts of one jointly trained model.
Funding snapshot$7.6MDisclosed seed round
MaturityPublished large-scale training research
Per-user weightsNo personal model by design
Read the dossier
Purpose & product
Company claim Keep advanced models under distributed control through protocol learning, with contributors owning only parts of a network. [PL2]
Reported Agora’s July 2026 paper reports an 8.6B-parameter model trained on 500B tokens over 40 days with 330 contributing nodes. These are author-reported experiment results, not independently reproduced benchmarks. [PL4]
Market & go-to-market
Analysis Initially researchers and GPU contributors; potentially organizations seeking shared control of model training and serving. Protocol participation and future inference economics are more evidenced than a mature enterprise sales business. [PL2] [PL5]
Founding & leadership
Company claim Early 2024, according to the seed announcement. [PL1]
- CEO
- Alexander Long ↗Cofounder / CEO [PL1]
- CTO
- Not separately verifiedNo current CTO title established
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Company claim $7.6M seed announced 19 March 2025, led by Union Square Ventures and CoinFund. No later confirmed round found. [PL1] [PL2]
Unverified Recurring revenue, commercial customer count, valuation and cash balance not established.
How the system learns
Reported Subspace Networks partition layers across nodes and compress inter-stage activations and backward signals in low-dimensional spaces. This is model parallelism, unlike a fleet of complete personal models exchanging user-specific knowledge. [PL3]
Company claim Protocol learning aims to prevent any single participant from possessing the whole model. Extraction resistance depends on the protocol and adversary assumptions; sharding alone is not proof against collusion. [PL2]
Implemented / publicly evidenced
Documented Public technical papers and a reported multi-contributor training run. [PL3] [PL4]
Roadmap / not yet established
Roadmap Scale open training / inference and contributor economics. A dated general-availability or revenue roadmap was not verified. [PL5]
What to verify next
Evaluate node churn, collusion and extraction assumptions, and compare end-to-end training efficiency including communication.
Profile last verified: · 5 cited sources · Official site / repository ↗
06 / Shared-model infrastructure
Nous Research
Two layers: distributed model training and a personal agent that builds reusable skills.
Funding snapshot$158MReported cumulative total, October 2026
MaturityOpen models + agent; business expansion
Per-user weightsNot automatic in Hermes Agent
Read the dossier
Purpose & product
Company claim Give people and businesses control of adaptable intelligence through open models and agent software. [NO1]
Documented Hermes models and the MIT-licensed Hermes Agent are public. The release archive states Hermes 4.3 Seed was post-trained on Psyche. Business offerings were announced in October 2026. [NO6] [NO7] [NO1]
Market & go-to-market
Analysis Open-model and agent adoption creates a developer funnel into hosted plans and business / enterprise deployments. Enterprise distribution is a newer focus, not proof of mature enterprise penetration. [NO1]
Founding & leadership
Company claim 2023 formation, stated by the CEO in October 2026. Earlier informal community activity is not the same as company formation. [NO1]
- CEO
- Dillon Rolnick ↗Current CEO [NO1] [NO4]
- CTO
- Jeffrey Quesnelle ↗Current CTO [NO4]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Reported $50M Series A, April 2025, chiefly Paradigm. [NO3]
Company claim $90M announced October 2026; participants include NVIDIA, M12, Samsung, Robot Ventures, USV, YC and Menlo. [NO1]
Reported Reported total: $158M; valuation: $1.5B. Earlier-round breakdown remains unverified. [NO2]
Unverified Current audited revenue, cash and paid-customer counts were not verified. Agent downloads, clones and token usage should not be interpreted as unique paying users.
How the system learns
Reported Psyche uses low-communication distributed optimization with peer coordination and verification. DeMo compresses momentum in a discrete-cosine-transform basis and exchanges selected components while retaining local state. The DisTrO / DeMo research line reduces synchronization traffic; it does not pool personal memories. [NO5]
Documented Hermes Agent persists context and generates reusable skills. That improvement mechanism does not, by itself, update the underlying model weights. Users may self-host and select a model provider. [NO7]
What to verify next
Separate the agent’s memory/skills metrics from model-training metrics; validate permissioning, reproducibility and enterprise deployment controls.
Profile last verified: · 8 cited sources · Official site / repository ↗
07 / Shared-model infrastructure
Prime Intellect
Open training infrastructure for organizations that want to own their optimization loop.
Funding snapshot>$150MCompany cumulative total; lower-bound bar
MaturityCommercial training / inference platform
Per-user weightsCustomer training, not default per-person
Read the dossier
Purpose & product
Company claim Make training, deployment and continual improvement of open models accessible outside major labs. [PR1]
Documented Lab, compute, inference and open research artifacts are available. On 13 October 2026, Prime announced the acquisition of Moondream to expand multimodal models and inference; deal terms were not disclosed. [PR3] [PR4]
Market & go-to-market
Analysis AI startups, model builders and enterprise agent teams. Open research and developer tooling feed paid compute, reinforcement learning, evaluation and inference services. [PR1]
Founding & leadership
Reported Founders began in late 2023; public company activity and seed financing followed in 2024. Legal incorporation date not verified. [PR10] [PR3]
- CEO
- Vincent Weisser ↗Cofounder / CEO [PR9]
- CTO
- Johannes Hagemann ↗Cofounder / CTO [PR8]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Company claim $5.5M seed in April 2024, co-led by Distributed Global and CoinFund; $15M funding announced February 2025, led by Founders Fund. [PR11] [PR12]
Company claim $130M Series A on 8 July 2026, led by Radical Ventures with NVIDIA Ventures, Intel Capital and Dell Technologies Capital. Company states total funding exceeds $150M. [PR1]
Documented Transaction counsel reports a $1B valuation for the Series A. [PR2]
Company claim July 2026 announcement reports more than 6,000 customers and over $100M in annualized revenue. This is an issuer run-rate claim, not audited annual revenue or a stated recurring-software ARR figure. [PR1]
How the system learns
Reported INTELLECT-1 uses DiLoCo-style local optimization with infrequent synchronization of model updates. INTELLECT-2 distributes reinforcement-learning rollouts, while policy training consumes their results. INTELLECT-3 extends the open-model program. This does not establish federated learning over private personal data. [PR5] [PR6] [PR7]
Analysis Customers can train models for their workflows and control resulting deployments, subject to model and service terms. A customer-owned optimization loop differs from every consumer running an independent evolving model.
Implemented / publicly evidenced
Documented Public model reports and a commercial training stack; the reported decentralized experiments should be assessed separately from ordinary hosted cloud jobs. [PR3] [PR5] [PR6]
Roadmap / not yet established
Roadmap Larger RL runs, long-horizon agents, automated research and continual production learning. These are explicit investment directions, not evidence that all models learn safely during production use. [PR1]
What to verify next
Request workload-specific quality/cost results, ownership terms, reproducible training recipes and robust controls against reward hacking.
Profile last verified: · 12 cited sources · Official site / repository ↗
08 / Shared-model infrastructure
Gensyn
Make AI computation auditable across hardware; connect it to information markets.
Funding snapshot$67.3MCalculated disclosed venture rounds; excludes token sale
MaturityVerifiable AI artifacts + market platform
Per-user weightsNot a per-user model product
Read the dossier
Purpose & product
Company claim Enable verifiable machine intelligence and an open economy for useful information. [GE3] [GE4]
Reported Delphi launched in April 2026 as an AI-settled information-market product. Market-creator and network fees form its stated commercial design; broad adoption is not independently established. [GE2]
Market & go-to-market
Analysis AI infrastructure builders and information-market creators / traders. This is now broader than a GPU marketplace, but remains adjacent to personal model ownership.
Founding & leadership
Documented 2020, per investor company record and founder history. [GE1]
- CEO
- Ben Fielding ↗Cofounder / CEO [GE1] [GE2]
- CTO
- Harry Grieve ↗Cofounder / CTO per investor record [GE1]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Company claim $1.1M pre-seed in 2021; $6.5M seed in 2022 led by Eden Block; $43M Series A in June 2023 led by a16z crypto. [GE5] [GE6]
Reported $16.7M venture round in October 2025, per April 2026 reporting. Known venture rounds sum to $67.3M; completeness is not assured. [GE2]
Reported Separate December 2025 token sale: $11.7M proceeds after refunds, versus $16.14M commitments. The reported $1B figure is fully diluted network value, not an equity valuation. [GE2]
Unverified Current audited revenue, cash and active paying-customer figures were not verified. Trading volume and token proceeds should not be treated as software revenue.
How the system learns
Documented REE / RepOps enforce repeatable arithmetic and execution across supported hardware. open-1b publishes training artifacts and hashes so others can replay steps. Its September 2026 release explicitly says the original training was centralized; auditing is decentralized. [GE3] [GE4]
Analysis Replay can establish that computation matches a declared process. It cannot establish that a model’s answer is true, its dataset is benign, or market incentives cannot be gamed.
Implemented / publicly evidenced
Documented open-1b checkpoints, training evidence and an audit harness were released; the issuer reports 1.61B total parameters and 400B training tokens. We did not rerun the audit. [GE4]
What to verify next
Distinguish deterministic execution, training provenance and output correctness; quantify audit coverage and verification overhead.
Profile last verified: · 6 cited sources · Official site / repository ↗
09 / Federated learning
Flower Labs
Practical infrastructure for learning across data that stays with its owners.
Funding snapshot$23.6MSum of two identified rounds; may not be lifetime total
MaturityOpen framework + enterprise platform
Per-user weightsSupported as a developer-built design
Read the dossier
Purpose & product
Company claim Train useful models on distributed, privately held data without first pooling all raw data centrally. [FL1]
Documented Open federated-learning framework, FlowerTune examples and enterprise / SuperGrid platform materials exist. Flower Intelligence began as a local/cloud inference preview; Endeavor 1.0 is an access-by-request model preview announced in September 2026. [FL4] [FL5] [FL6]
Market & go-to-market
Analysis Organizations with data silos, including healthcare, finance and industrial teams. Open-source developer adoption supports enterprise deployment and infrastructure services. [FL1]
Founding & leadership
Documented Flower Labs formed in 2023; the open-source Flower framework dates to 2020. [FL2] [FL3]
- CEO
- Daniel J. Beutel ↗Cofounder / CEO; LinkedIn destination from YC [FL1] [FL7]
- CTO
- Not separately listedTechnical cofounder Nicholas Lane is CSO, not CTO [FL1]
- Chief scientist
- Nicholas (Nic) Lane ↗Cofounder / CSO; LinkedIn destination from YC [FL1] [FL7]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Company claim $3.6M pre-seed in 2023, led by First Spark Ventures; $20M Series A on 15 February 2024, led by Felicis. Identified rounds total $23.6M. [FL1]
Unverified Current revenue, valuation, cash and paid-user figures were not verified. Open-source usage is not equivalent to paying enterprise adoption.
How the system learns
Documented Clients train locally; a strategy aggregates updates. The FlowerTune example uses LoRA / parameter-efficient LLM adaptation and FedAvg. Its documented simulation is an example, not proof of a global production consumer deployment. [FL4]
Analysis Federation can retain raw data at each participant. The deployment owner chooses aggregation, personalization and safeguards; a coordinator may remain central. Differential privacy and secure aggregation must be explicitly configured and evaluated. [FL4] [ME1]
Implemented / publicly evidenced
Documented Runnable framework and federated LLM example, with local/cloud inference tooling on a separate product track. [FL4] [FL5]
Roadmap / not yet established
Roadmap Broader private collaborative AI, models and enterprise infrastructure. No dated rollout of an independently evolving model for every consumer was verified. [FL6]
What to verify next
A strong implementation candidate: test non-identical client data, device costs, privacy budgets and personalization retention after aggregation.
Profile last verified: · 7 cited sources · Official site / repository ↗
10 / Federated learning
TensorOpera / FedML
Federated-learning infrastructure alongside a broader model and agent business.
Funding snapshot$13.2MUndated company seed figure; differs from dated $11.5M
MaturityFederated toolkit + commercial agent platform
Per-user weightsDeveloper-enabled, not established for every user
Read the dossier
Purpose & product
Company claim Make model development, deployment and agent creation economical for developers and enterprises. [TE1]
Documented FedML / TensorOpera federated documentation, model serving and agent tools are public. Teamily’s July 2026 launch extends the product surface toward collaborative consumer / team agents. [TE3] [TE4] [TE7]
Market & go-to-market
Analysis Enterprise model and agent builders, plus teams using agent products. Developer onboarding and hosted training / serving support commercial infrastructure sales. [TE1] [TE3]
Founding & leadership
Company claim February 2022 company formation. The FedML research library predates the company, with a 2020 paper. [TE1] [TE6]
- CEO
- Aiden Chaoyang He ↗Current cofounder / CEO [TE1]
- CTO
- Not separately listedDo not recycle older CEO / CTO assignments [TE1]
- Executive chairman
- Salman AvestimehrCofounder; current title on company page; LinkedIn URL not verified [TE1]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Company claim July 2023 release reports $11.5M seed in total: $4.3M earlier financing plus $7.2M additional funding, led by Camford. Do not add $4.3M again. [TE2]
Company claim Current company biography says $13.2M seed. Timing and reconciliation of the extra $1.7M were not verified. Related ChainOpera financing is not included. [TE1]
Unverified Current audited revenue, valuation and cash balance not established. Platform-registration and ecosystem-user claims are not comparable to paid customers.
How the system learns
Documented FedML supports coordinated training across data holders. Separately, TensorOpera Router selects models for requests, and agent products manage context and tools. Routing or persistent agent memory should not be mistaken for a new personal model trained on every interaction. [TE4] [TE5]
Analysis Privacy and model ownership depend on the chosen federated or hosted deployment. No reviewed evidence establishes universal, ongoing per-person fine-tuning across its consumer products.
What to verify next
Choose the product layer first: federated SDK, hosted model platform and Teamily have different learning and data-control properties.
Profile last verified: · 7 cited sources · Official site / repository ↗
11 / Intelligence ecosystems
Sentient
Open agent improvement and model ownership, without requiring personal weight updates.
Funding snapshot$85MDisclosed seed; separate token entry excluded
MaturityOpen research + composable agent ecosystem
Per-user weightsNot required by EvoSkill
Read the dossier
Purpose & product
Company claim Advance open reasoning and community participation in the development and economics of AI. [SN2]
Documented Current public work includes EvoSkill, Arena, OML and the GRID ecosystem. GRID’s launch was reported in September 2025. Product existence does not establish decentralized training of personal models. [SN2] [SN3] [SN4]
Market & go-to-market
Analysis Open-source agent developers, researchers and enterprise evaluation teams. Community challenges and reusable research attract contributors; sustainable commercial conversion remains less visible. [SN2]
Founding & leadership
Reported 2024; January founding reported by a company directory. Exact incorporation date not verified. [SN6]
- CEO
- Not verifiedNo reliable current CEO designation established
- CTO
- Not verifiedNo reliable current CTO designation established
- Cofounder / researcher
- Himanshu Tyagi ↗Founding announcement also names Sandeep Nailwal, Pramod Viswanath and Sensys [SN1]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Company claim $85M seed announced July 2024, led by Founders Fund, Pantera and Framework. [SN1]
Reported A token-transparency database lists $7.6M on 19 January 2026. Primary proceeds reconciliation was not recovered; keep this separate and unverified. [SN5]
Unverified Audited revenue, cash, valuation and paid-user counts not verified. Ecosystem partner totals do not establish recurring revenue.
How the system learns
Documented EvoSkill uses evaluation and failed agent trajectories to propose reusable skills. It improves the agent’s instructions / tools without requiring model training. OML addresses model ownership and usage accountability; a composable agent network is a different mechanism from federated weight exchange. [SN2] [SN3]
Analysis Open artifacts improve inspectability, while platform rules and incentives still shape participation. Benchmark gains do not prove broad continual learning or robust personal-data privacy.
Implemented / publicly evidenced
Documented EvoSkill code and public research / evaluation surfaces. [SN2] [SN3]
Roadmap / not yet established
Roadmap Eldros is listed as coming soon with an expected 2026 release; the page still presents a waitlist at the cutoff. Broader AGI ambitions remain goals. [SN2]
What to verify next
Audit evaluation design and reward gaming; separate skill evolution, model fingerprinting and model-weight training in demos.
Profile last verified: · 6 cited sources · Official site / repository ↗
12 / Intelligence ecosystems
Bittensor
An incentive network for useful machine outputs, not a single jointly trained model.
Funding snapshotNot comparableProtocol ecosystem; no unified corporate funding ledger
MaturityOperating protocol / subnet ecosystem
Per-user weightsNot a universal protocol feature
Read the dossier
Purpose & product
Company claim Reward useful machine intelligence through a network of specialized markets. [BI2]
Documented Public protocol documentation defines miners, validators, subnet owners and staking. This report does not audit live subnet quality, throughput or profitability. [BI2]
Market & go-to-market
Analysis Subnet builders, model / service operators and applications consuming their outputs. Economic rewards attract supply, while demand and commercial utility must be assessed subnet by subnet. [BI2]
Founding & leadership
Documented Network launched November 2021, per the 2022 paper. This is a protocol-launch date, not verified incorporation of a single company. [BI1]
- CEO
- Not applicable to the protocolCurrent foundation corporate officers not verified
- CTO
- Not applicable to the protocolSubnet teams have their own leadership
- Cofounder
- Jacob Steeves ↗Paper also names cofounder Ala Shaabana [BI1]
Roles and professional links last checked 13 October 2026. Unlisted links were not guessed; LinkedIn may require sign-in.
Funding & business evidence
Not comparable No single corporate round ledger represents the protocol. TAO purchases, token market capitalization and subnet financing are not interchangeable with startup capital raised.
Unverified Network-wide business revenue and foundation financials were not reconciled. Token emissions reward participation; they are not necessarily customer demand.
How the system learns
Documented Miners supply outputs or services; validators score contributions; consensus and incentives allocate rewards. Each subnet defines its task. The protocol does not automatically merge independently trained weights into one improving global model. [BI2] [BI3]
Analysis Control is distributed across subnet rules, validators and stake. Incentive alignment, evaluator quality and concentration matter as much as the number of nodes.
Implemented / publicly evidenced
Documented A documented operating protocol and tooling ecosystem; individual subnet capability requires separate diligence. [BI1] [BI2]
Roadmap / not yet established
Unverified No single verified, dated roadmap covers all subnets. Future model-learning mechanisms depend on individual teams and protocol evolution.
What to verify next
Inspect a specific subnet’s task, anti-gaming design, real buyers and net economics before attributing intelligence gains to the network.
Profile last verified: · 3 cited sources · Official site / repository ↗
04 / Evidence & method
Know what a claim can support.
Public-source research through 13 October 2026. The report distinguishes record verification from validation of a system’s performance.
Evidence rules
Documented
A filing, public artifact, product interface, named role or implementation description was located. No implied independent performance audit.
Company claim
An attributable issuer statement, including fundraising and product claims. Press releases remain issuer evidence even when syndicated.
Reported
Secondary reporting, or explicitly attributed research results. The original system or transaction was not independently reproduced.
Roadmap
A future direction, expected launch, proposed mechanism or gated objective. It is not credited as shipped.
Unverified
Missing, conflicting or insufficiently supported information. This means unresolved, not false or nonexistent.
Analysis
Our interpretation of the evidence, including architecture families, go-to-market implications and diligence priorities.
Learning is not one mechanism
- Memory / RAG: store and retrieve context; weights can remain unchanged.
- Skills: improve procedures, prompts or tools; this need not retrain the model.
- LoRA / fine-tuning: optimize model parameters or small adapters using data and a loss function.
- Federated learning: train across data holders and aggregate updates; it may still use a central coordinator.
- Distributed training: spread a shared training job across machines; it does not imply user data sovereignty.
- Incentive networks: reward contributions; usefulness depends on evaluation and economic design.
Local data does not automatically mean private learning. Model updates can leak training information. Privacy claims need a threat model, privacy accounting and evidence about what each participant can observe. [ME1]
Financial and leadership discipline
Round amounts are not added when they may overlap. Token commitments, net proceeds, network valuations, revenue run rates and equity financing are kept distinct. An older financial disclosure is labeled with its period.
CEO and CTO are explicit fields. Missing titles are not inferred from a cofounder’s technical background. Professional LinkedIn URLs are included only where matched to the person; some pages require sign-in. Flower’s technical cofounder is labeled CSO, and PIN’s Regan Peng has a visible title change.
Limits of the snapshot
Sources were inspected for relevance and attribution, with preference for filings, official documentation, repositories, research papers and issuer announcements. No company was contacted. We did not execute training jobs, audit smart contracts, verify private cap tables or test paid product accounts.
Undated pages and professional profiles can change. “Last verified” records the research date, not proof that every statement on a living page was historically true. No post-cutoff claims are intentionally included. This is a curated competitive set, not an exhaustive industry census.
Scope choices. Flower and TensorOpera are included as practical federated-learning enablers; Sentient and Bittensor as alternative coordination and ownership mechanisms. Hylon is retained because its stated design is directly relevant, with evidence gaps explicit. SELF OS means the selflabs/SELF-OS project, not similarly named companies. Generic memory frameworks, broad personal-assistant products and compute-only marketplaces are not separately profiled.
Source register / 78 recordsOpen all source URLs + dates
- PA1Regulatory filingHuman AI Labs — SEC annual report, FY2023 ↗
Published: 2024-04-24
https://www.sec.gov/Archives/edgar/data/1835770/000183577024000001/humanailabs.pdf
Last verified: - PA2Company / primaryPersonal AI — company profile and current leadership ↗
Published: Undated / living page
https://www.linkedin.com/company/personalai/
Last verified: - PA3Issuer fundraising pagePersonal AI — Republic campaign and founder updates ↗
Published: Campaign ended 2024-04-01; subsequent updates
https://republic.com/personal-ai
Last verified: - PA4Secondary filing indexHuman AI Labs — reported Form D history ↗
Published: Includes 2025-12-18 entry
https://www.noairlab.ai/companies/human-ai-labs-inc
Last verified: - PA5Company / primaryPersonal AI — MODEL-2 architecture and release ↗
Published: 2024-01-09
https://www.personal.ai/pi-ai/ces-2024-personal-ai-model-2
Last verified: - PA6Technical documentationPersonal AI — persona context and memory retrieval ↗
Published: Undated / living page
https://www.personal.ai/pai-academy/chat-prompting-and-carrying-context-with-a-persona
Last verified: - PA7Issuer press releasePersonal AI — Comcast AI Grid trials ↗
Published: 2026-03-17
https://www.globenewswire.com/news-release/2026/03/17/3257092/0/en/Personal-AI-s-Memory-Based-Small-Language-Models-Deliver-Hyper-Personalized-Experiences-on-Comcast-s-AI-Grid-Powered-by-NVIDIA.html
Last verified: - PA8Syndicated issuer announcementPersonal AI — $7.8M seed-capital announcement ↗
Published: 2023-01-10
https://venturebeat.com/business/personal-ai-has-raised-7-8-million-in-seed-capital-to-build-its-personal-language-model-sets-out-to-revolutionize-human-to-human-conversations-with-personal-ais-prepares-for-series-a-in-2023
Last verified: - PI1Company / primaryPIN AI — $10M pre-seed and founding team ↗
Published: 2024-09-09
https://www.pinai.com/post/pin-ai-team-from-ethereum-google-brain-stanford-and-mit-secures-10m-from-hack-vc-a16z-csx-and
Last verified: - PI2Technical documentationPIN AI — open platform technical design ↗
Published: 2025-01
https://www.pinai.com/post/pin-ai-the-open-platform-for-personal-ai
Last verified: - PI3Research paperGuardian of Data: training privacy-preserving personal assistants ↗
Published: 2025-02-24
https://arxiv.org/abs/2502.18527
Last verified: - PI4Code / artifactPIN AI — GOD model repository ↗
Published: Undated / living page
https://github.com/PIN-AI/God-Model
Last verified: - PI5Product pagePIN AI — current app distribution page ↗
Published: Undated / living page
https://www.pinai.com/product
Last verified: - PI6Self-authored professional bioRegan Peng — current president biography ↗
Published: 2026 profile
https://councils.forbes.com/profile/Regan-Peng-Co-Founder-President-PIN-AI/cd6fa4de-e22c-4b4f-b802-7e880f3ed1c8
Last verified: - PI7Third-party event announcementAWS event post — Regan Peng described as CTO ↗
Published: 2025-05
https://www.linkedin.com/posts/bradfein_awsstartups-consensus2025-activity-7326268170367897601-qiLf
Last verified: - HY1Company / primaryHylon — current positioning and node beta ↗
Published: 2026 working draft
https://hylon.ai
Last verified: - HY2Technical documentationOrbVPN — Hylon implementation boundaries ↗
Published: Undated / living page
https://www.orbvpn.com/en/docs/tutorials/hylon-ai/decentralised-ai
Last verified: - HY3Technical documentationHylon — five-layer technical architecture ↗
Published: 2026 working draft
https://hylon.ai/technology/
Last verified: - HY4Company whitepaperHylon — whitepaper and gated roadmap ↗
Published: 2026 working draft
https://hylon.ai/whitepaper/
Last verified: - SE1Code / artifactSELF-OS — community repository and explicit limits ↗
Published: Undated / living page
https://github.com/selflabs/SELF-OS
Last verified: - SE2Code / artifactSELF-OS — open-source boundary ↗
Published: Undated / living page
https://github.com/selflabs/SELF-OS/blob/main/docs/OPEN_SOURCE_BOUNDARY.md
Last verified: - SE3Self-authored professional bioMilan Cheeks — SELF Labs professional profile ↗
Published: Undated / living page
https://www.linkedin.com/in/milan-cheeks-61828099
Last verified: - SE4Secondary organization directorySELF Labs — Milan Cheeks listed as CEO ↗
Published: Undated / living page
https://theorg.com/org/self-labs-inc/person/milan-cheeks
Last verified: - PL1Issuer press releasePluralis — founding and $7.6M seed ↗
Published: 2025-03-19
https://www.globenewswire.com/news-release/2025/03/19/3045635/0/en/Pluralis-Research-Pioneers-Protocol-Learning-to-Scale-Decentralized-AI-Announces-7-6M-Seed-Round-Led-by-USV-and-CoinFund.html
Last verified: - PL2Investor / primaryUSV — investment in Pluralis ↗
Published: 2025-03-19
https://www.usv.com/writing/pluralis-towards-actually-open-ai
Last verified: - PL3Research paperSubspace Networks: scaling distributed training ↗
Published: 2025-06-02; revised 2026-07-09
https://arxiv.org/abs/2506.01260
Last verified: - PL4Research paperAgora — large-scale protocol learning ↗
Published: 2026-07-14
https://arxiv.org/abs/2607.13332
Last verified: - PL5Company / primaryPluralis — current research and network ↗
Published: Undated / living page
https://pluralis.ai
Last verified: - NO1Company / primaryNous — October 2026 fundraise and Hermes strategy ↗
Published: 2026-10
https://nousresearch.com/a-note-on-our-fundraise
Last verified: - NO2Independent reportingNous — Series B total and valuation reporting ↗
Published: 2026-10-07
https://techcrunch.com/2026/10/07/nous-research-confirms-it-hit-1-5b-valuation-launches-ai-agents-for-business-users/
Last verified: - NO3Independent reportingNous — $50M Paradigm financing ↗
Published: 2025-04-25
https://www.fortune.com/crypto/2025/04/25/paradigm-nous-research-crypto-ai-venture-capital-deepseek-openai-blockchain/
Last verified: - NO4Company / primaryNous — current CEO and CTO named together ↗
Published: 2026
https://www.linkedin.com/posts/nousresearch_our-ceo-dillon-rolnick-and-cto-jeffrey-quesnelle-activity-7468309035247714304-mUV0
Last verified: - NO5Technical documentationNous — Psyche architecture ↗
Published: 2025-05
https://nousresearch.com/nous-psyche
Last verified: - NO6Release recordNous — model and agent releases ↗
Published: Undated / living page
https://nousresearch.com/releases
Last verified: - NO7Technical documentationHermes Agent documentation ↗
Published: Undated / living page
https://hermes-agent.nousresearch.com/docs
Last verified: - NO8Company technical roadmapPsyche — future directions ↗
Published: 2025-10-09
https://forum.nousresearch.com/t/psyche-future-directions/273
Last verified: - PR1Company / primaryPrime Intellect — Series A, business metrics and roadmap ↗
Published: 2026-07-08
https://www.primeintellect.ai/blog/series-a
Last verified: - PR2Transaction participantCooley — Prime Intellect Series A legal counsel announcement ↗
Published: 2026-07-08
https://www.cooley.com/news/coverage/2026/2026-07-08-prime-intellect-raises-%24130-million-series-a
Last verified: - PR3Company / primaryPrime Intellect — company announcements and earlier rounds ↗
Published: 2024–2026 archive
https://www.primeintellect.ai/blog
Last verified: - PR4Company / primaryPrime Intellect — Moondream acquisition ↗
Published: 2026-10-13
https://www.primeintellect.ai/blog/moondream
Last verified: - PR5Research paperINTELLECT-1 technical report ↗
Published: 2024-12-02
https://arxiv.org/abs/2412.01152
Last verified: - PR6Research paperINTELLECT-2 technical report ↗
Published: 2025-05-12
https://arxiv.org/abs/2505.07291
Last verified: - PR7Research paperINTELLECT-3 technical report ↗
Published: 2025-12
https://arxiv.org/abs/2512.16144
Last verified: - PR8Self-authored professional bioJohannes Hagemann — cofounder / CTO biography ↗
Published: Undated / living page
https://github.com/JohannesHa
Last verified: - PR9Self-authored professional bioVincent Weisser — Prime Intellect CEO profile ↗
Published: Undated / living page
https://www.linkedin.com/in/vincentweisser
Last verified: - PR10Founder event biographyPrime Intellect founders — late-2023 origin ↗
Published: Undated / living page
https://luma.com/p9rmebmm
Last verified: - PR11Company / primaryPrime Intellect — $15M funding announcement ↗
Published: 2025-02-28
https://www.primeintellect.ai/blog/fundraise
Last verified: - PR12Investor / primarySeed investor — Prime Intellect $5.5M financing ↗
Published: 2024-04
https://www.linkedin.com/posts/ben-fielding_pleased-to-back-johannes-hagemann-and-vincent-activity-7189330587462086656-CloV
Last verified: - GE1Investor / primaryEntrepreneurs First — Gensyn founding, CEO and CTO ↗
Published: Undated / living page
https://www.joinef.com/portfolio/?company=gensyn
Last verified: - GE2Independent reportingGensyn — Delphi launch and financing update ↗
Published: 2026-04-22
https://www.theblock.co/news/business/2026-04-22-a16z-crypto-gensyn-delphi-ai-settled-information-markets-platform-398419
Last verified: - GE3Technical documentationGensyn — REE architecture ↗
Published: 2026-03-17
https://www.gensyn.ai/news/ree
Last verified: - GE4Company technical releaseGensyn — open-1b auditable training release ↗
Published: 2026-09-15
https://www.gensyn.ai/news/introducing-open-1b-auditable-training
Last verified: - GE5Founder announcementGensyn — $6.5M seed and $1.1M pre-seed ↗
Published: 2022-03
https://www.linkedin.com/posts/harry-grieve-81427771_today-we-announce-our-65m-seed-round-activity-6911585117786271744-s92r
Last verified: - GE6Founder announcementGensyn — $43M Series A ↗
Published: 2023-06
https://www.linkedin.com/posts/harry-grieve-81427771_today-were-announcing-our-43m-series-a-activity-7073799813590319104-ZYVW
Last verified: - FL1Company / primaryFlower Labs — Series A and leadership ↗
Published: 2024-02-15
https://flower.ai/blog/2024-02-15-announcing-series-a
Last verified: - FL2Company / primaryFlower Labs — company formation ↗
Published: 2023-03-08
https://flower.ai/blog/2023-03-08-flower-labs
Last verified: - FL3Research paperFlower — initial framework paper ↗
Published: 2020-07
https://arxiv.org/abs/2007.14390
Last verified: - FL4Code / technical documentationFlowerTune — federated LLM fine-tuning example ↗
Published: Undated / living page
https://flower.ai/docs/examples/flowertune-llm-general-nlp.html
Last verified: - FL5Company technical releaseFlower Intelligence preview ↗
Published: 2025-03-12
https://flower.ai/blog/2025-03-12-flower-intelligence
Last verified: - FL6Company technical releaseFlower — Endeavor 1.0 model announcement ↗
Published: 2026-09-01
https://flower.ai/blog/2026-09-01-introducing-endeavor-1.0
Last verified: - FL7Investor / primaryY Combinator — Flower founders and profile links ↗
Published: Undated / living page
https://www.ycombinator.com/companies/flower
Last verified: - TE1Company / primaryTensorOpera — founding, leadership and $13.2M figure ↗
Published: Undated / living page
https://tensoropera.ai/about/who-we-are
Last verified: - TE2Issuer press releaseFedML — $11.5M combined seed announcement ↗
Published: 2023-07-19
https://www.businesswire.com/news/home/20230719154444/en/FedML-Closes-%2411.5-Million-Seed-Round-to-Help-Companies-Build-and-Train-Custom-Generative-AI-Models
Last verified: - TE3Product pageTensorOpera — products and platform ↗
Published: Undated / living page
https://tensoropera.ai
Last verified: - TE4Technical documentationTensorOpera — federated learning documentation ↗
Published: Undated / living page
https://docs.tensoropera.ai/federate/getting_started
Last verified: - TE5Research paperTensorOpera Router paper ↗
Published: 2024-08
https://arxiv.org/abs/2408.12320
Last verified: - TE6Research paperFedML research library paper ↗
Published: 2020-07
https://arxiv.org/abs/2007.13518
Last verified: - TE7Syndicated issuer announcementTeamily launch announcement ↗
Published: 2026-07-13
https://press.todayan.com/newsRead.php?no=1038445
Last verified: - SN1Company / primarySentient — seed financing and cofounders ↗
Published: 2024-07-02
https://www.linkedin.com/posts/sentientagi_we-are-thrilled-to-announce-sentients-85m-activity-7213909618270371841-XDXT
Last verified: - SN2Company / primarySentient — research, products and Eldros roadmap ↗
Published: Undated / living page
https://www.sentient.xyz
Last verified: - SN3Code / artifactEvoSkill — reusable agent skill evolution ↗
Published: Undated / living page
https://github.com/sentient-agi/EvoSkill
Last verified: - SN4Independent reportingSentient — GRID launch ↗
Published: 2025-09-16
https://www.theblock.co/news/business/2025-09-16-founders-fund-backed-sentient-launches-open-source-composable-ai-ecosystem-370518
Last verified: - SN5Secondary token-finance databaseSentient — token transparency entry ↗
Published: Includes 2026-01-19 entry
https://blockworks.com/token-transparency/filing/sentient
Last verified: - SN6Secondary company directorySentient — founding-date directory entry ↗
Published: Undated / living page
https://seedtable.com/companies/sentient
Last verified: - BI1Research paperBittensor — 2022 paper and network origins ↗
Published: 2022
https://bittensor.com/pdfs/academia/NeurIPS_DAO_Workshop_2022_3_3.pdf
Last verified: - BI2Technical documentationBittensor — current network documentation ↗
Published: Undated / living page
https://www.bittensor.com/docs
Last verified: - BI3Protocol paperBittensor — utility consensus mechanism ↗
Published: Undated / living page
https://bittensor.com/pdfs/consensus_v2/PoS_Utility_Consensus.pdf
Last verified: - ME1Research paperDeep Leakage from Gradients ↗
Published: 2019-06-21; revised 2019-12-19
https://arxiv.org/abs/1906.08935
Last verified: